So no one cares I’m sure. But I thought I’d share my experience and thoughts before I go.
I think what the AI was actually doing in my case, or at least what it felt like, was predicting where my ftp will be at the end of the block. And then progressively ramping me to meet its prediction. It’s early guesses were tragically wrong, so wrong I corrected them one time, but then went with it. My future ftp prediction was considerable and stable until the last week.
At which point it saw something I would call a benchmark, basically I was scheduled for an endurance ride and then it did basic standard math to predict my future ftp. That first crack in its prediction was a slightly higher hr on that endurance ride, what some might call hr drift, that effectively has to be the reason… The final workout was a vo2max which I am not currently trained for, and not particularly good at, at which point it predicted my future ftp to be less than the current one because I straight up failed. 
. Again I have to assume Its just doing math. Before you say anything yeah I know it’s not doing math, I’m saying that in a statistical set of people it is bending to a natural distribution. I should be capable of doing x% of ftp for y minutes z times. And if I can’t my ftp is wrong. Now of course it could also just be forcing my ftp lower so it can find training for me to do. But then that feels a bit awkward too. It doesn’t need a narrative, but sadly I sort of do. 
The end result? I’d have to guess I went from about 312 to about 330 in 8 weeks. I’ll do a ramp test just for kicks to see the end result
As a side note late last year I went from 312 to 336 or so only doing zone 2, literally (As a bit of a meme test) it was not nearly as time efficient as trainerRoad and my high end was terrible at the end, only doing endurance+ every day will kill your top end. Shoulda seen the sea of yellow. Heh. Anyway. I gave it a shot. Still a great app and I love the zwift integration. 
